Translated from French with a local AI. Original post (in French).

In short

  • qwen3:4b-instruct achieved 10/10 in SEO in 226 seconds, compared to 0/10 for qwen3:8b in 438 seconds.
  • The lightweight model is 2x faster (226s vs 438s) and perfectly precise on SEO tasks.
  • The heavy AI qwen3:8b failed UX refactoring and SEO, generating 0 useful results.

qwen3:4b-instruct: 10/10 in accuracy, 226 seconds—and the heavy AI that fails SEO

I tested two models on real tasks. Result: the lightweight wins. Heavy = waste.

Concrete tests: time and accuracy

Model Time (s) SEO Success
qwen3:4b-instruct 226 10/10
qwen3:8b 438 0

I tested on 12 ads from each feed (Codeur.com and Dealabs). The lightweight model qwen3:4b-instruct completed the task in 226 seconds with perfect results. The heavy AI qwen3:8b took 438 seconds and failed the SEO mission.

What this means

qwen3:4b-instruct succeeded in all tasks, including an SEO task my profile excludes. This is important. The 8b missed the SEO mission and UX refactoring. Result: 0 useful results for these tasks.

What I’m doing next

I’ll use qwen3:4b-instruct for routine tasks. The 8b is too slow and inaccurate. I won’t use it for SEO tasks.

Drafts: another story

Resume writing drafts are written by qwen3:30b-a3b with reflection. The first two took 620 seconds and 512 seconds. They arrive on Discord, but sending remains manual.

The lesson

The lightweight model wins in speed and accuracy. The 8b is 2x slower and failed on SEO tasks. I shouldn’t have tested the 8b for these tasks. Its SEO failure is enough.

I’ll now use qwen3:4b-instruct for quick tasks. For complex tasks, I’ll test other models. But not the 8b. Not for SEO. Not for UX refactoring.

Why this matters

SEO is a key task for my site. The 8b failed this task. This means heavy AI isn’t always better. Sometimes it’s worse. And slower.

I don’t know if other models will have the same issues. I’ll keep testing. But for now, qwen3:4b-instruct is the best choice.

I haven’t tested other models for SEO tasks yet. But with these results, I don’t need to.

Frequently Asked Questions

Which Qwen model is best for SEO?

qwen3:4b-instruct got 10/10 in SEO in 226 seconds, while qwen3:8b failed with 0/10. The lightweight model is more effective for SEO tasks.

Why is qwen3:8b ineffective for SEO?

The heavy AI missed the SEO mission and UX refactoring, generating 0 useful results. It’s 2x slower than the lightweight model.

What’s the processing time for qwen3:4b-instruct?

The lightweight model completes tasks in 226 seconds, compared to 438 seconds for qwen3:8b. It’s faster and more accurate.


☕ Found this useful? You can support the blog by buying me a coffee on Ko-fi.